MLflow Model Serving - From Training to Production

Explore MLflow Model Serving in this 37-minute Data + AI Summit Europe 2020 Meetup presentation by Andre Mesarovic, Resident Solutions Architect at Databricks. Dive into the world of hosting machine learning models as REST endpoints with automatic updates, enabling data science teams to manage the entire lifecycle of real-time ML models from training to production. Learn about scoring models with MLflow, both online using the MLflow scoring server and offline with Apache Spark, as well as custom model deployment and scoring techniques. Gain insights into the MLflow life cycle, Databricks Model Serving, and recent features like Torch Serving. Discover the intricacies of MLflow Saved Model Format, Flavors, and Code, and witness a practical demonstration using Onix. This comprehensive talk covers essential vocabulary, deployment overviews, and various scoring methods, providing a solid foundation for understanding and implementing MLflow Model Serving in your data science projects.

Intro
Welcome
Offline vs Online
Vocabulary
MLflow
Spark
Online Scoring
MLflow Scoring Server
Deployment
Overview
Scoring Server
CSV Request
Sagemaker
Flask vs Spark
MLeap
MLflow Life Cycle
Databricks Model Serving
Model Serving Overview
Model Registry UI
Recent Resources
Plumbing Plugin
New Features
Torch Serving
Example
MLflow Saved Model Format
MLflow Saved Model Flavor
MLflow Saved Model Code
Launching Docker Container
Demo
Onix
Questions